基于图割算法的摄影测量点云面向对象分类方法

摘要

针对摄影测量点云分类问题,提出一种基于图割算法的面向对象分类方法.由于摄影测量点云质量相对较差,文中首先基于区域增长分割思想,提出一种新的点云体素生成方法,将摄影测量点云聚为不同的对象;然后将不同的对象作为节点,将分类问题建模为一个多标记问题,引入图割算法优化获取邻域内平滑一致的分类结果.利用两组摄影测量点云数据进行实验,正确率分别为87.3% 和88.7%,比基于单点的分类方法分别提高2% 和2.6%.%Aiming at the classification problem of photogrammetric point cloud, an object-based classification method using a graph cuts algorithm is proposed.Due to the relative poor quality of the photogrammetric point cloud,this paper firstly proposes a new supervoxels generating method to segment the photogrammetric point cloud into different objects based on the idea of region growing.After that,the classification problem is modeled as a multi-label task,w hile different objects are treated as nodes and the graph cuts algorithm is used to obtain a consistent classification result.The correct rates on two groups of data sets are 87.3% and 88.7%,w hich are 2% and 2.6% higher than the result in using point-based classification method.